Joel is a senior staff research scientist at Google DeepMind. He obtained his PhD from MIT where he studied computational neuroscience and machine learning with Tomaso Poggio. Joel was one of the first researchers to join DeepMind, starting as an intern in 2010, and then joining full time after finishing his PhD in 2013. He is interested in reverse engineering human biological and cultural evolution to inform the development of artificial intelligence that is simultaneously human-like and human-compatible. In particular, Joel believes that theories of cooperation from fields like cultural evolution and institutional economics can be fruitfully applied to inform the development of ethical and effective artificial intelligence technology.

Melting Pot

Melting Pot is a suite of test scenarios and a protocol for evaluating multi-agent reinforcement learning algorithms. Melting Pot assesses generalization to novel social situations involving both familiar and unfamiliar individuals. It covers a broad range of social concepts including cooperation, competition, reciprocation, trust, stubbornness, ownership, exchange, fair division, conventions, norms, negotiation, and so forth.

Melting Pot is an open source project. It is available on github [here].  Melting Pot is under active development and contributions are very welcome. The usefulness of any evaluation suite is a function of its coverage. Right now Melting Pot contains more than 250 test scenarios and covers a wide range of topics. But not all concepts of interest to the cooperative artificial intelligence and multi-agent reinforcement learning research communities are yet represented. Useful contributions to Melting Pot may take form either as new learning environments or as new background populations (sets of pre-trained agents), each typically constituting a new social adaptive challenge. Over time we hope Melting Pot will grow into a comprehensive benchmark incorporating tests for machine analogs to most of the important social-cognitive capacities, representations, and motivations underpinning human intelligence, cooperation, and culture.

Publications


Selected (old) conference abstracts